A pipeline that goes quiet, and an inbox that never empties
Admissions is a pipeline, not an event. An applicant submits a form, and from that moment the file needs a transcript, a reference, an identity document, a deposit — and most files are incomplete on first submission. Somebody has to notice what's missing, chase it, and check again once it arrives. When a team is sized for an average week, the applicant who doesn't reply to the first reminder often just goes quiet, and nobody notices until the file is stale and the decision window has closed. That applicant didn't decide against your institution. Nobody had the hours to find out.
Student services runs the opposite problem. The questions aren't hard, they're just constant — a locked portal account, a transcript request, a fee-deadline date, what the withdrawal or refund process actually involves, where a form lives. Every one of those has an answer already sitting in a handbook or a policy document somewhere. Answering it from scratch, over email, a hundred times a term, is a poor use of a skilled person's attention, and it's exactly the kind of load that pushes the harder questions — the ones that genuinely need a person — to the bottom of the queue.
What the agent does, day to day
A custom AI agent scoped to your institution's actual workflows typically covers some combination of the following. It does not decide who gets admitted, does not grade anything, and does not stand in for a member of staff — it moves the process along and puts a finished picture in front of a person for the call that's theirs to make.
- Applicant pipeline management. Every open file is tracked against what it's still missing — transcript, reference, identity document, deposit — rather than living as a stalled row in a spreadsheet nobody has reopened in a week.
- Document and reference chasing. Reminders go out to the applicant or the referring school on a set cadence, in a tone that matches your institution's voice, until the item arrives or the file needs a person's attention.
- Deposit and enrollment-fee follow-up. An accepted offer with no deposit yet is followed up on schedule, and a pattern that looks like real hesitation — not just a missed reminder — gets flagged to a counsellor instead of another automatic nudge.
- Nudge-or-human triage. The agent decides, file by file, whether the next step is a routine reminder or a flag to a human — a document that's arrived twice with mismatched names, an applicant who's gone silent after three attempts, anything that reads as more than a paperwork gap.
- Student services inbox triage. Incoming messages are read, categorized, and matched against what your institution's own policies actually say, so a routine question gets a routine, accurate answer instead of waiting behind everything else in the queue.
- Alumni and fundraising segmentation. Alumni and donor records are segmented by engagement, giving history, or lapsed contact, so outreach goes to the right list instead of one message sent to everyone.
- Outreach drafting. Campaign and stewardship messages are drafted in your institution's voice and queued for a person to approve, not sent on the agent's own judgment.
- Internal staff HR and onboarding requests. Leave requests, document collection, and the repeated steps of bringing on new or seasonal staff get moved along automatically, the same way the agent works staff onboarding for any organization with a recurring hiring cycle.
Walkthrough: an applicant who goes quiet after the deposit request
An offer is accepted and a deposit is requested with a stated deadline. The agent logs the offer, sets the follow-up cadence, and sends the first reminder a few days before the deadline through whichever channel your institution actually uses — email, portal message, or both. No response arrives. It sends a second reminder closer to the deadline, checking first whether the applicant has opened or clicked the previous message, because a reminder to someone who's never opened the last three needs a different approach than a nudge to someone actively engaging.
If the deadline passes with no deposit and no response, the agent doesn't keep sending identical reminders into the void. It checks the file for signals — has the applicant engaged with any other communication, is there an open question sitting in an inbox somewhere, has a sibling or counselor conversation already happened — and either escalates to an admissions counselor with that context attached, or extends the window automatically if your institution's policy allows a standard grace period. Either way, a person makes the call on what happens to the applicant. The agent's job was making sure nobody found out three weeks late that the file had gone cold.
Walkthrough: triaging the student services inbox against your own policies
A message arrives asking how to request an official transcript before a job application deadline. The agent reads it, matches the request against your institution's actual transcript policy — processing time, fee, delivery method — and replies with the correct, specific answer, because the answer comes from your documented policy rather than a guess. It logs the interaction so there's a record of what the student was told.
The next message in the queue asks about withdrawing mid-term for a reason the student describes as a personal crisis. The agent reads the framing, recognizes this isn't a routine "what's the process" question, and routes it directly to a student services staff member rather than replying with the standard withdrawal procedure. Anything that reads as a welfare or safeguarding concern is built to escalate immediately and without hesitation — that routing rule sits above every other instruction the agent has, not behind it. The measure of a good inbox agent here isn't how many messages it closes. It's that the one message that actually needed a person reached one, fast.
Walkthrough: segmenting an alumni list and drafting the outreach
Your development office wants to run a spring appeal targeted at alumni from a specific graduating cohort who haven't given in the last few cycles but have a history of event attendance. The agent pulls the segment from your donor and alumni database against those criteria, checks for any do-not-contact flags, and assembles the list.
It then drafts the outreach message in your institution's established voice, referencing the cohort's specific graduation year and the event history where relevant, and queues the draft for your development team to review and approve before anything sends. Once responses and pledges come in, it logs them back to the donor record as a proper entry rather than a loose note in someone's inbox. The relationship with a major donor or a long-standing alumnus still runs through your development team — the agent handles the segmentation and the first draft, not the conversation.
Not the strategy engagement
It's worth being clear about what this page isn't. If your institution is still working out which administrative workflow to tackle first, how to draft an acceptable-use policy, or how to separate an operational fix from an academic-integrity question that belongs to your teaching staff, that's AI strategy for education — an advisory engagement that produces a plan and a sequence, not a running system. This page describes the build: a working agent operating inside your admissions pipeline, your inbox, and your donor records once you already know which workflow is worth fixing. Institutions often start with the strategy engagement and move into a build like this one once the first project is named — but plenty come to us already knowing exactly which queue is costing them the most hours, and start here directly. Either way, this system sits inside the broader picture of what Calfy builds for schools and universities — one agent, scoped to the workflow that matters most right now.
What this agent will never decide
An agent that touches admissions and student records has real limits, and we hold them deliberately rather than as a footnote. Calfy does not build systems that grade student work, make admissions decisions, or replace a teacher, tutor, or advisor. Assessment and enrollment decisions require a person's judgment, and a system that quietly took that judgment away would be a liability to your institution long before it became a convenience. What the agent does is move the paperwork and the routine questions along so the people making those calls are looking at a complete file and a clear picture — not a faster path to a decision made by software.
The same discipline applies to anything that isn't paperwork. A question that reads as a genuine welfare or safeguarding concern — a student in distress, a disclosure, anything outside a routine administrative request — is routed straight to a person immediately, with no attempt by the agent to handle it, soften it, or hold it for the next business day.
Student data and access control
Anything that touches an applicant file or a student record has to be built around that fact from the start, not added on before launch. Student and applicant data is sensitive by nature — academic history, financial details, and sometimes information tied to welfare or wellbeing — and your institution carries its own obligations around how that information is handled, which your own data protection or compliance lead is best placed to define for your context.
In practice, that means the agent gets the narrowest access that lets it do its job: read-only where reading is enough, scoped credentials rather than a shared login, and detailed logging of what it touched and why. It means a clear boundary between what the agent may surface to a person and what it may act on directly, and it means the escalation rules described above sit ahead of everything else the agent does, not behind it.
Where it connects into your systems
Admissions and student information systems. Most platforms expose an API or a scheduled data feed, which is usually enough to read applicant files and write status updates without disrupting your record of truth. Where a system is older or closed, there's typically still a workable path in — an export job, a database view, or a portal the agent can operate the way a person would.
Email and the student services inbox. The agent reads and drafts from the same inbox your team already uses, rather than requiring students or applicants to learn a new channel.
Donor and alumni CRM. Segmentation, outreach drafts, and logged responses run against whichever alumni or fundraising platform your development office already runs.
HR and onboarding systems. Staff and onboarding requests move through your existing HR platform or ticketing tool, with the agent handling the repeated steps rather than replacing the system itself.
Escalation. Every workflow has a defined edge — an applicant whose file raises a question beyond a paperwork gap, an inbox message that reads as a welfare concern, a donor relationship that needs a person's judgment. When the agent reaches that edge, it stops and hands the case to someone with the full context attached, rather than guessing at what happens next. For the underlying mechanics of what separates an agent from a simple script, see the AI agent glossary entry.